CN-122020051-A - Ocean space-time target change enhancement recognition method based on space difference regulation and control

CN122020051ACN 122020051 ACN122020051 ACN 122020051ACN-122020051-A

Abstract

The invention discloses a space-time target change enhancement recognition method based on space difference regulation, which quantitatively depicts the contribution degree of each space unit in an overall change structure in a space-time field of ocean, regulates and controls a data representation structure based on the contribution difference, so that the tensor decomposition process is guided to more intensively and stably show the change process related to a target analysis task in a decomposition result while the expression capacity of the overall structure is maintained, and the analysis method suitable for unstructured ocean space-time field data is constructed on the premise of not depending on a regular grid assumption and a complete observation condition, so that space-time change components with definite physical significance can be stably extracted under the condition of lacking a measurement value and geographical boundary constraint, and the automatic screening and recognition of the decomposition result are realized by introducing target-oriented evaluation reference.

Inventors

LI DONGSHUANG
PAN LIMING
HUANG CHENGQIANG
LUO WEN
YU ZHAOYUAN
YUAN LINWANG

Assignees

南京师范大学

Dates

Publication Date: 20260512
Application Date: 20260202

Claims (8)

1. A method for enhancing and identifying ocean space-time target change based on space difference regulation and control is characterized by comprising the following steps of (1) Constructing a multidimensional tensor and a mask tensor consistent with the tensor dimension according to unstructured ocean space-time field data, and determining an observation tensor of mask constraint; (2) Under the constraint condition of a mask tensor, carrying out low-rank tensor decomposition on the observation tensor, and determining a space-time response component, wherein the space-time response component comprises a time factor matrix, a space factor matrix and a variable factor matrix; (3) Calculating the importance of the space unit according to the space factor matrix; (4) Constructing a space difference regulating factor based on the importance of the space unit, and carrying out weighting construction on the observation tensor in the step (1) in the space dimension through the space difference regulating factor to obtain a reconstructed observation tensor; (5) Defining a target analysis task, and constructing an evaluation reference for evaluating the validity of the region identification result according to the target analysis task; (6) And calculating a consistency index between the new time factor and the time reference of the target change process according to the evaluation reference and the new time factor matrix, and outputting an enhanced recognition result of the target change process.
2. The method for enhanced recognition of marine space-time target changes based on spatial differential regulation of claim 1, wherein in step (1), a multidimensional tensor is used Is that ; Wherein, the Representing a real set, T representing a time dimension, N representing the number of ocean space units, and V representing an ocean environment element or variable dimension; Mask tensor consistent with tensor dimension Is that ; Wherein, the mask element is 1 to represent effective ocean observation, and the mask element is 0 to represent missing value, ineffective observation or position which does not meet geographical boundary constraint; mask constrained observation tensor Is that ; Wherein, the Representing the elements in the corresponding positions for multiplication.
3. The method for enhancing and identifying the change of the marine space-time target based on the space difference regulation and control according to claim 2, wherein the step (2) is specifically as follows: by minimizing the reconstruction error under the constraint of the mask tensor, the following optimization problem is solved: ; Wherein, the Represent the first The number of time factor vectors is a function of the time factor vector, Representing the corresponding spatial response vector(s), Representing a variable response vector; The vector outer product is represented by the sum of the vectors, Is the Frobenius norm; obtaining low rank decomposition results meeting observation constraints : ; Wherein, the Each representing an independent pattern of variation in a time-space-variation direction, The time factor matrix is used for describing the whole time evolution mode; the space factor matrix is a space factor matrix, and the row vector of the space factor matrix corresponds to the response characteristic of each space unit in the main change structure; Is a variable factor matrix and is used for describing the coupling relation among multiple variables.
4. The method for enhanced recognition of marine space-time target changes based on spatial differential regulation of claim 3, wherein in step (3), spatial unit importance is determined Is that ; Wherein, the Represent the first The degree of participation of the individual spatial units, R, represents the number of independent change patterns extracted in the tensor decomposition.
5. The method for enhanced identification of changes in marine space-time objects based on spatial differential control of claim 4, wherein in step (4), the spatial differential control factor Is that ; Wherein, the For regulating index parameters, for controlling spatial difference regulating intensity, and ; Reconstructed observation tensor Is that ; Wherein, the Representing low rank decomposition results At the position of And the value of (c) above.
6. The method for enhancing and identifying marine space-time target change based on space difference regulation as claimed in claim 5, wherein in step (6), tensor decomposition is performed again on the reconstructed observation tensor to obtain a new space-time component Is that ; Wherein, the Represents the first obtained when tensor decomposition is performed on the observed tensor subjected to space difference regulation A time factor vector for describing the evolution characteristic of the corresponding variation component in the time dimension; Representing a corresponding spatial factor vector describing the spatial response distribution characteristics of the variation component over each marine space cell; Representing the corresponding variable factor vector for characterizing the response of the variable component in different marine environment variable dimensions.
7. The method for enhanced identification of marine space-time target changes based on spatial differential control according to claim 6, wherein in step (6), a consistency index between a time factor and a target change process time reference is obtained Is that ; Wherein, the Representing a correlation metric function, S representing an evaluation reference; ; ; ; Wherein, the And Respectively representing the sequence means.
8. The method for enhancing and identifying the change of the marine space-time target based on the space difference regulation and control according to claim 7, wherein the enhanced identification result of the change process of the target is output Is that 。

Description

Ocean space-time target change enhancement recognition method based on space difference regulation and control Technical Field The invention relates to the technical field of ocean data processing and ocean information, in particular to an ocean space-time target change enhancement recognition method based on space difference regulation. Background The analysis method for the ocean space-time field data mainly comprises a principal component analysis method, a singular value decomposition method and a multidimensional data analysis method based on tensor decomposition. For example, principal component analysis is widely used for dominant variation analysis of marine elements such as sea level height, sea surface temperature, sea surface air pressure, etc., by decomposing a space-time covariance structure of a marine variable to extract a set of orthogonal time modes and space modes. In recent years, with the increasing demand for multivariate, multidimensional data analysis, typical tensor decomposition methods (such as CP decomposition and turner decomposition) have been introduced into the field of marine science to characterize the coupling structure between time-space-variables. However, the above existing methods generally implicitly or explicitly assume that different spatial units have equal weight contributions in the decomposition process, i.e. the impact weight of each spatial position on the temporal modality in the tensor decomposition or matrix decomposition process is determined passively by the data itself, and lack an active regulatory mechanism for the difference of spatial contributions. In a marine spatiotemporal field, there is often a significant difference in the roles of different spatial regions in the physical process, with some critical regions having a dominant effect on a particular marine phenomenon or climate index, whereas a large background variation or noise region may occupy a larger energy specific gravity in the decomposition result, resulting in the targeted variation process being dispersed or weakened in multiple components. In addition, the prior art generally relies on pre-processing steps such as interpolation, re-meshing, etc. when processing unstructured marine observation data to meet the analysis requirements of a regular grid or complete data. Such processing may introduce additional errors, and may also tend to attenuate the true physical signal in areas of severe lack of detection or complex geographic boundaries, adversely affecting subsequent decomposition results. The ocean space-time field tensor decomposition or matrix decomposition analysis method based on the equal weight space assumption has the core aim of extracting dominant variation modes from ocean space-time data and is used for representing the integral space-time evolution characteristics. The method generally takes regular or re-gridded ocean observation data as input, and obtains a group of time modes and space modes through a linear decomposition model. The prior art generally pre-processes raw marine observation data, including temporal interpolation, spatial resampling, and missing value filling, to convert unstructured or incomplete observation data into a regular time-space grid data structure. The processed data is generally organized in a two-dimensional matrix or in a three-dimensional tensor form, on the basis of which the prior art uses principal component analysis or a typical tensor decomposition method to decompose the data. In the prior art, each spatial unit is generally regarded as having the same weight in the decomposition process, and its spatial factor is automatically determined in the overall energy-optimal sense directly by the decomposition algorithm, without distinguishing or regulating the relative importance of different spatial regions in a specific ocean process. The time factors resulting from the decomposition are typically ranked by energy magnitude or variance contribution and interpreted empirically by an analyst to infer potential physical processes or climate modalities. The prior art mainly shows a unidirectional processing flow of data input-equal weight decomposition-modal output, namely, from pretreated ocean space-time field data, a plurality of time and space modes are obtained through one-time decomposition, and then the modes are directly output as analysis results, and a feedback regulation or result screening mechanism aiming at a specific target analysis task is lacked. In general, the prior art takes equal weight tensor decomposition or matrix decomposition as a core in the realization structure, and can effectively extract the overall dominant change characteristics of the ocean space-time field, but under the application scene that the space contribution has obvious difference and the target change process is dominant by a local area, the problems that the target signal is dispersed in a plurality of components, covered by background change o